MCP / Models / Voice, music and sound
MiniMax Music 3 MCP: full songs with vocals up to 5 minutes
MiniMax Music 3 writes and sings a full song of up to five minutes from a style prompt and your lyrics, as a stereo WAV file. Run it from Claude Code, Codex, Cursor or any MCP client through one FlyMy.AI MCP, or call its API.

$ YOUR NEXT RUN
Use MiniMax Music 3 in your assistant
Paste this prompt into Claude Code, Codex, Cursor or any agent. It reads the guide and helps you connect.
Read https://flymy.ai/skill.md and https://flymy.ai/mcp/models/minimax-music-3.md. Help me connect the FlyMy.AI MCP to this assistant, then run MiniMax Music 3 (flymyai/minimax-music-3) for this task: Write a short indie pop song about a summer road trip with MiniMax Music 3, with a verse and a chorus.
What MiniMax Music 3 does
MiniMax Music 3 (MiniMax-Music3) is MiniMax's open-weights music model. Describe the music in prompt, required: genre, mood, vocals, instruments and arrangement; for precise control write a structured caption with the genre, BPM, key, how the mood builds, the voice and a section-by-section arrangement. Write the words in lyrics, required, with a structure tag on a line of its own before each part ([intro], [verse], [pre-chorus], [chorus], [post-chorus], [bridge], [instrumental], [solo] or [outro]); words written on the same line after a tag are dropped. duration is the longest the song may be, 1 to 300 seconds (60 by default), and the model may stop earlier. seed repeats a result, num_inference_steps (1 to 100, 30 by default) trades speed for quality and guidance_scale (0 to 20, 1.7 by default) sets how closely the music follows the prompt. Outputs: audio, a 44.1 kHz 16-bit stereo WAV file (about 10 MB a minute), seed and duration, its real length in seconds. In our tests a 30-second song took about a minute and a quarter to make; a full-length song takes longer.
It is billed on what the run uses, by the seconds of music made, and the price shown before the run is the longest the song can be, held on your balance; you are charged what the run cost. It makes one song per request. It does not take a reference song or a melody, clone a voice or make speech or sound effects.
MiniMax Music 3 inputs
Every input of flymyai/minimax-music-3 as its live schema lists them. Returns audio (array), seed (string), duration (string).
| Input | Type, values and default |
|---|---|
# prompt | string required Describe the music: genre, mood, vocals, instruments and arrangement; for precise control write a structured caption (genre, BPM, key, how the mood builds, the voice, then each section) |
# lyrics | string required The words to sing. Put a structure tag on a line of its own before each part: [intro], [verse], [pre-chorus], [chorus], [post-chorus], [bridge], [instrumental], [solo] or [outro]; words on the same line after a tag are dropped. For example a [verse] line, the verse on the lines below, then a [chorus] line and the chorus |
# duration | number optional Allowed: 1 to 300 Default: 60 The longest the song may be, in seconds, 1-300 (60 by default): the model may stop earlier. $0.002 per second |
# seed | integer optional Allowed: 0 to 2147483647 Seed for a repeatable result |
# num_inference_steps | integer optional Allowed: 1 to 100 Default: 30 Quality steps per 8-second part of the song, 1-100 (30 by default): more steps improve the quality and take longer |
# guidance_scale | number optional Allowed: 0 to 20 Default: 1.7 How closely the music follows the prompt, 0-20 (1.7 by default) |
Ask your agent
Requests to paste as they are, once the FlyMy.AI MCP is connected.
flymyai/minimax-music-3flymyai/minimax-music-3flymyai/minimax-music-3MiniMax Music 3 price
$0.002 per second of duration. A run at the default settings costs $0.12. A run costs $0.002 to $0.60, depending on duration. Billed to your FlyMy.AI balance; no seats, no subscription and no minimum.
| Setting | Price |
|---|---|
| Any | $0.002 per second |
Call MiniMax Music 3 from code
The same model, from an agent or from your own code. Keys stay in your FlyMy.AI account.
From an agent, through the MCP
The agent calls get_model_schema, then run_model with:
{
"endpoint_id": "flymyai/minimax-music-3",
"input": {
"prompt": "Warm lo-fi beat with soft piano and light vinyl crackle",
"lyrics": "[verse]\nCity lights and a slow train home"
}
}From your code, over HTTP
Form fields, not JSON; files go as uploads or public URLs. The answer is a prediction_id; read the result at https://api.flymy.ai/api/v1/flymyai/minimax-music-3/predict/async/result/?request_id=<prediction_id>.
curl -X POST https://api.flymy.ai/api/v1/flymyai/minimax-music-3/predict/async/ \
-H "X-API-KEY: $FLYMYAI_API_KEY" \
-F 'prompt=Warm lo-fi beat with soft piano and light vinyl crackle' \
-F 'lyrics=[verse]
City lights and a slow train home'Questions about MiniMax Music 3
Do I need a MiniMax API key for MiniMax Music 3?
No. MiniMax Music 3 runs on FlyMy.AI's account, so you need no MiniMax key or plan: sign in to FlyMy.AI once and each run is billed to your FlyMy.AI balance.
How much does MiniMax Music 3 cost?
$0.002 per second of duration. There is no subscription or minimum; the price comes from FlyMy.AI billing and was checked on 2026-11-18.
How do I use MiniMax Music 3 in Claude Code, Codex or Cursor?
Add the FlyMy.AI MCP with one command and sign in, then ask in plain words, for example: "Write a short indie pop song about a summer road trip with MiniMax Music 3, with a verse and a chorus." Your agent reads the inputs with get_model_schema and runs flymyai/minimax-music-3 with run_model.
What does MiniMax Music 3 need as input?
MiniMax Music 3 requires prompt and lyrics; it also takes 4 optional inputs, each listed on this page with its allowed values and default.
Can I call MiniMax Music 3 from my own code?
Yes. Send form fields to https://api.flymy.ai/api/v1/flymyai/minimax-music-3/predict/async/ with your FlyMy.AI API key in the X-API-KEY header.
Go to the source
From the FlyMy.AI model catalog of 2026-11-18: inputs from the model's live schema, prices from FlyMy.AI billing. Requests are examples to paste, not recorded runs.





